Deep HST/NICMOS H ( F160W ) band observations of the z = 1.96 quasar pair Q 1634+267A , B reveal no signs of a lens galaxy to a 1 \sigma threshold of \simeq 22.5 mag .
The minimum luminosity for a normal lens galaxy would be a 6 L ^ { * } galaxy at z \simeq 0.5 , which is 650 times greater than our detection threshold .
Our observation constrains the infrared mass-to-light ratio of any putative , early-type , lens galaxy to ( M / L ) _ { H } \gtrsim 690 h _ { 65 } ( 1200 h _ { 65 } ) for \Omega _ { 0 } = 0.1 ( 1.0 ) and H _ { 0 } = 65 h _ { 65 } ~ { } \hbox { km~ { } s } ^ { -1 } ~ { } \hbox { Mpc } ^ { -1 } .
We would expect to detect a galaxy somewhere in the field because of the very strong Mg ii absorption lines at z = 1.1262 in the Q 1634+267 A spectrum , but the HST H-band , I-band ( F785LP ) and V-band ( F555W ) images require that any associated galaxy be very under-luminous \lesssim 0.1 L ^ { * } _ { H } ( 1.0 L ^ { * } _ { I } ) if it lies within \lesssim 40 h _ { 65 } ^ { -1 } ( 100 h _ { 65 } ^ { -1 } ) kpc from Q 1634+267 A and B .

While the large image separation ( 3 \farcs 85 ) and the lack of a lens galaxy strongly favor interpreting Q 1634+267A , B as a binary quasar system , the spectral similarity remains a puzzle .
We estimate that at most 0.06 % of randomly selected quasar pairs would have spectra as similar to each other as the spectra of Q 1634+267 A and B .
Moreover , spectral similarities observed for the 14 known quasar pairs are significantly greater than would be expected for an equivalent sample of randomly selected field quasars .
Depending on how strictly we define similarity , we estimate that only 0.01–3 % of randomly drawn samples of 14 quasar pairs would have as many similar pairs as the observational sample .